How a conversion lift study works
The platform takes the people it would otherwise show your ads to and divides them at random. One group is eligible to see the campaign; the other is held back and shown nothing from you. Both groups are then followed to see how many convert. Because the division happens inside the ad system, on the same audience, at the same moment, the only systematic difference between the groups is the advertising itself.
The output is expressed as lift: the extra conversions among the exposed group over and above what the held-back group produced on its own. That is a different quantity from the conversions your campaign normally reports, which are simply the ones the platform could attribute to a click or a view.
Why conversion lift studies matter
Attributed conversions and caused conversions are not the same thing, and the gap is usually widest exactly where the budget is largest — remarketing, branded search, broad prospecting on social. A lift study is the platform’s own answer to that question, and it is far easier to arrange than building your own holdout test, because the randomisation, the audience and the reporting are all handled for you.
It also gives a budget conversation something firmer to stand on. “This campaign is credited with these sales” invites argument about the attribution model. “Sales were higher among people who could see it than among people who could not” is a claim about cause, and it is much harder to wave away.
Where conversion lift studies go wrong
Availability is the first limit. Platforms restrict these studies by spend, by market and by account type, and the conditions change, so treat access as something to confirm rather than assume. A study that cannot reach a readable result should not be started at all.
The second problem is interpretation. An inconclusive result is often read as proof the campaign works, on the grounds that nothing was disproved. It is not. It usually means the effect, if there is one, was smaller than the design could see. And the answer belongs to the campaign, audience and period that were tested — not to the channel in general, and not automatically to next quarter.
How to use one
Run it on a decision you would genuinely act on. Choose the campaign whose value you are least sure of, agree in advance what result would lead you to cut it, keep it or expand it, and then leave the settings alone for the duration. Changing bids or creative mid-study introduces a second variable into a design that can only handle one.
Afterwards, set the lift against the conversions the platform reported for the same window. The relationship between the two is a useful correction to carry into planning, and it makes far better evidence than any attribution model on its own. Make sure the conversion being measured is one that reflects real business value, because a lift study inherits whatever your conversion tracking setup is already counting.
